1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext2/inode.c 4 * 5 * Copyright (C) 1992, 1993, 1994, 1995 6 * Remy Card (card@masi.ibp.fr) 7 * Laboratoire MASI - Institut Blaise Pascal 8 * Universite Pierre et Marie Curie (Paris VI) 9 * 10 * from 11 * 12 * linux/fs/minix/inode.c 13 * 14 * Copyright (C) 1991, 1992 Linus Torvalds 15 * 16 * Goal-directed block allocation by Stephen Tweedie 17 * (sct@dcs.ed.ac.uk), 1993, 1998 18 * Big-endian to little-endian byte-swapping/bitmaps by 19 * David S. Miller (davem@caip.rutgers.edu), 1995 20 * 64-bit file support on 64-bit platforms by Jakub Jelinek 21 * (jj@sunsite.ms.mff.cuni.cz) 22 * 23 * Assorted race fixes, rewrite of ext2_get_block() by Al Viro, 2000 24 */ 25 26 #include <linux/time.h> 27 #include <linux/highuid.h> 28 #include <linux/pagemap.h> 29 #include <linux/blkdev.h> 30 #include <linux/quotaops.h> 31 #include <linux/writeback.h> 32 #include <linux/buffer_head.h> 33 #include <linux/mpage.h> 34 #include <linux/fiemap.h> 35 #include <linux/iomap.h> 36 #include <linux/namei.h> 37 #include <linux/uio.h> 38 #include "ext2.h" 39 #include "acl.h" 40 #include "xattr.h" 41 42 /* 43 * Test whether an inode is a fast symlink. 44 */ 45 static inline int ext2_inode_is_fast_symlink(struct inode *inode) 46 { 47 int ea_blocks = EXT2_I(inode)->i_file_acl ? 48 (inode->i_sb->s_blocksize >> 9) : 0; 49 50 return (S_ISLNK(inode->i_mode) && 51 inode->i_blocks - ea_blocks == 0); 52 } 53 54 static void ext2_truncate_blocks(struct inode *inode, loff_t offset); 55 56 void ext2_write_failed(struct address_space *mapping, loff_t to) 57 { 58 struct inode *inode = mapping->host; 59 60 if (to > inode->i_size) { 61 truncate_pagecache(inode, inode->i_size); 62 ext2_truncate_blocks(inode, inode->i_size); 63 } 64 } 65 66 /* 67 * Called at the last iput() if i_nlink is zero. 68 */ 69 void ext2_evict_inode(struct inode * inode) 70 { 71 struct ext2_block_alloc_info *rsv; 72 int want_delete = 0; 73 74 if (!inode->i_nlink && !is_bad_inode(inode)) { 75 want_delete = 1; 76 dquot_initialize(inode); 77 } else { 78 dquot_drop(inode); 79 } 80 81 truncate_inode_pages_final(&inode->i_data); 82 83 if (want_delete) { 84 sb_start_intwrite(inode->i_sb); 85 /* set dtime */ 86 EXT2_I(inode)->i_dtime = ktime_get_real_seconds(); 87 mark_inode_dirty(inode); 88 sync_inode_metadata(inode, inode_needs_sync(inode)); 89 /* truncate to 0 */ 90 inode->i_size = 0; 91 if (inode->i_blocks) 92 ext2_truncate_blocks(inode, 0); 93 ext2_xattr_delete_inode(inode); 94 } else { 95 mmb_sync(&EXT2_I(inode)->i_metadata_bhs); 96 } 97 mmb_invalidate(&EXT2_I(inode)->i_metadata_bhs); 98 clear_inode(inode); 99 100 ext2_discard_reservation(inode); 101 rsv = EXT2_I(inode)->i_block_alloc_info; 102 EXT2_I(inode)->i_block_alloc_info = NULL; 103 if (unlikely(rsv)) 104 kfree(rsv); 105 106 if (want_delete) { 107 ext2_free_inode(inode); 108 sb_end_intwrite(inode->i_sb); 109 } 110 } 111 112 typedef struct { 113 __le32 *p; 114 __le32 key; 115 struct buffer_head *bh; 116 } Indirect; 117 118 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) 119 { 120 p->key = *(p->p = v); 121 p->bh = bh; 122 } 123 124 static inline int verify_chain(Indirect *from, Indirect *to) 125 { 126 while (from <= to && from->key == *from->p) 127 from++; 128 return (from > to); 129 } 130 131 /** 132 * ext2_block_to_path - parse the block number into array of offsets 133 * @inode: inode in question (we are only interested in its superblock) 134 * @i_block: block number to be parsed 135 * @offsets: array to store the offsets in 136 * @boundary: set this non-zero if the referred-to block is likely to be 137 * followed (on disk) by an indirect block. 138 * To store the locations of file's data ext2 uses a data structure common 139 * for UNIX filesystems - tree of pointers anchored in the inode, with 140 * data blocks at leaves and indirect blocks in intermediate nodes. 141 * This function translates the block number into path in that tree - 142 * return value is the path length and @offsets[n] is the offset of 143 * pointer to (n+1)th node in the nth one. If @block is out of range 144 * (negative or too large) warning is printed and zero returned. 145 * 146 * Note: function doesn't find node addresses, so no IO is needed. All 147 * we need to know is the capacity of indirect blocks (taken from the 148 * inode->i_sb). 149 */ 150 151 /* 152 * Portability note: the last comparison (check that we fit into triple 153 * indirect block) is spelled differently, because otherwise on an 154 * architecture with 32-bit longs and 8Kb pages we might get into trouble 155 * if our filesystem had 8Kb blocks. We might use long long, but that would 156 * kill us on x86. Oh, well, at least the sign propagation does not matter - 157 * i_block would have to be negative in the very beginning, so we would not 158 * get there at all. 159 */ 160 161 static int ext2_block_to_path(struct inode *inode, 162 long i_block, int offsets[4], int *boundary) 163 { 164 int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb); 165 int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb); 166 const long direct_blocks = EXT2_NDIR_BLOCKS, 167 indirect_blocks = ptrs, 168 double_blocks = (1 << (ptrs_bits * 2)); 169 int n = 0; 170 int final = 0; 171 172 if (i_block < 0) { 173 ext2_msg(inode->i_sb, KERN_WARNING, 174 "warning: %s: block < 0", __func__); 175 } else if (i_block < direct_blocks) { 176 offsets[n++] = i_block; 177 final = direct_blocks; 178 } else if ( (i_block -= direct_blocks) < indirect_blocks) { 179 offsets[n++] = EXT2_IND_BLOCK; 180 offsets[n++] = i_block; 181 final = ptrs; 182 } else if ((i_block -= indirect_blocks) < double_blocks) { 183 offsets[n++] = EXT2_DIND_BLOCK; 184 offsets[n++] = i_block >> ptrs_bits; 185 offsets[n++] = i_block & (ptrs - 1); 186 final = ptrs; 187 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { 188 offsets[n++] = EXT2_TIND_BLOCK; 189 offsets[n++] = i_block >> (ptrs_bits * 2); 190 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); 191 offsets[n++] = i_block & (ptrs - 1); 192 final = ptrs; 193 } else { 194 ext2_msg(inode->i_sb, KERN_WARNING, 195 "warning: %s: block is too big", __func__); 196 } 197 if (boundary) 198 *boundary = final - 1 - (i_block & (ptrs - 1)); 199 200 return n; 201 } 202 203 /** 204 * ext2_get_branch - read the chain of indirect blocks leading to data 205 * @inode: inode in question 206 * @depth: depth of the chain (1 - direct pointer, etc.) 207 * @offsets: offsets of pointers in inode/indirect blocks 208 * @chain: place to store the result 209 * @err: here we store the error value 210 * 211 * Function fills the array of triples <key, p, bh> and returns %NULL 212 * if everything went OK or the pointer to the last filled triple 213 * (incomplete one) otherwise. Upon the return chain[i].key contains 214 * the number of (i+1)-th block in the chain (as it is stored in memory, 215 * i.e. little-endian 32-bit), chain[i].p contains the address of that 216 * number (it points into struct inode for i==0 and into the bh->b_data 217 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect 218 * block for i>0 and NULL for i==0. In other words, it holds the block 219 * numbers of the chain, addresses they were taken from (and where we can 220 * verify that chain did not change) and buffer_heads hosting these 221 * numbers. 222 * 223 * Function stops when it stumbles upon zero pointer (absent block) 224 * (pointer to last triple returned, *@err == 0) 225 * or when it gets an IO error reading an indirect block 226 * (ditto, *@err == -EIO) 227 * or when it notices that chain had been changed while it was reading 228 * (ditto, *@err == -EAGAIN) 229 * or when it reads all @depth-1 indirect blocks successfully and finds 230 * the whole chain, all way to the data (returns %NULL, *err == 0). 231 */ 232 static Indirect *ext2_get_branch(struct inode *inode, 233 int depth, 234 int *offsets, 235 Indirect chain[4], 236 int *err) 237 { 238 struct super_block *sb = inode->i_sb; 239 Indirect *p = chain; 240 struct buffer_head *bh; 241 242 *err = 0; 243 /* i_data is not going away, no lock needed */ 244 add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets); 245 if (!p->key) 246 goto no_block; 247 while (--depth) { 248 bh = sb_bread(sb, le32_to_cpu(p->key)); 249 if (!bh) 250 goto failure; 251 read_lock(&EXT2_I(inode)->i_meta_lock); 252 if (!verify_chain(chain, p)) 253 goto changed; 254 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets); 255 read_unlock(&EXT2_I(inode)->i_meta_lock); 256 if (!p->key) 257 goto no_block; 258 } 259 return NULL; 260 261 changed: 262 read_unlock(&EXT2_I(inode)->i_meta_lock); 263 brelse(bh); 264 *err = -EAGAIN; 265 goto no_block; 266 failure: 267 *err = -EIO; 268 no_block: 269 return p; 270 } 271 272 /** 273 * ext2_find_near - find a place for allocation with sufficient locality 274 * @inode: owner 275 * @ind: descriptor of indirect block. 276 * 277 * This function returns the preferred place for block allocation. 278 * It is used when heuristic for sequential allocation fails. 279 * Rules are: 280 * + if there is a block to the left of our position - allocate near it. 281 * + if pointer will live in indirect block - allocate near that block. 282 * + if pointer will live in inode - allocate in the same cylinder group. 283 * 284 * In the latter case we colour the starting block by the callers PID to 285 * prevent it from clashing with concurrent allocations for a different inode 286 * in the same block group. The PID is used here so that functionally related 287 * files will be close-by on-disk. 288 * 289 * Caller must make sure that @ind is valid and will stay that way. 290 */ 291 292 static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind) 293 { 294 struct ext2_inode_info *ei = EXT2_I(inode); 295 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data; 296 __le32 *p; 297 ext2_fsblk_t bg_start; 298 ext2_fsblk_t colour; 299 300 /* Try to find previous block */ 301 for (p = ind->p - 1; p >= start; p--) 302 if (*p) 303 return le32_to_cpu(*p); 304 305 /* No such thing, so let's try location of indirect block */ 306 if (ind->bh) 307 return ind->bh->b_blocknr; 308 309 /* 310 * It is going to be referred from inode itself? OK, just put it into 311 * the same cylinder group then. 312 */ 313 bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group); 314 colour = (current->pid % 16) * 315 (EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16); 316 return bg_start + colour; 317 } 318 319 /** 320 * ext2_find_goal - find a preferred place for allocation. 321 * @inode: owner 322 * @block: block we want 323 * @partial: pointer to the last triple within a chain 324 * 325 * Returns preferred place for a block (the goal). 326 */ 327 328 static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block, 329 Indirect *partial) 330 { 331 struct ext2_block_alloc_info *block_i; 332 333 block_i = EXT2_I(inode)->i_block_alloc_info; 334 335 /* 336 * try the heuristic for sequential allocation, 337 * failing that at least try to get decent locality. 338 */ 339 if (block_i && (block == block_i->last_alloc_logical_block + 1) 340 && (block_i->last_alloc_physical_block != 0)) { 341 return block_i->last_alloc_physical_block + 1; 342 } 343 344 return ext2_find_near(inode, partial); 345 } 346 347 /** 348 * ext2_blks_to_allocate: Look up the block map and count the number 349 * of direct blocks need to be allocated for the given branch. 350 * 351 * @branch: chain of indirect blocks 352 * @k: number of blocks need for indirect blocks 353 * @blks: number of data blocks to be mapped. 354 * @blocks_to_boundary: the offset in the indirect block 355 * 356 * return the number of direct blocks to allocate. 357 */ 358 static int 359 ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks, 360 int blocks_to_boundary) 361 { 362 unsigned long count = 0; 363 364 /* 365 * Simple case, [t,d]Indirect block(s) has not allocated yet 366 * then it's clear blocks on that path have not allocated 367 */ 368 if (k > 0) { 369 /* right now don't hanel cross boundary allocation */ 370 if (blks < blocks_to_boundary + 1) 371 count += blks; 372 else 373 count += blocks_to_boundary + 1; 374 return count; 375 } 376 377 count++; 378 while (count < blks && count <= blocks_to_boundary 379 && le32_to_cpu(*(branch[0].p + count)) == 0) { 380 count++; 381 } 382 return count; 383 } 384 385 /** 386 * ext2_alloc_blocks: Allocate multiple blocks needed for a branch. 387 * @inode: Owner. 388 * @goal: Preferred place for allocation. 389 * @indirect_blks: The number of blocks needed to allocate for indirect blocks. 390 * @blks: The number of blocks need to allocate for direct blocks. 391 * @new_blocks: On return it will store the new block numbers for 392 * the indirect blocks(if needed) and the first direct block. 393 * @err: Error pointer. 394 * 395 * Return: Number of blocks allocated. 396 */ 397 static int ext2_alloc_blocks(struct inode *inode, 398 ext2_fsblk_t goal, int indirect_blks, int blks, 399 ext2_fsblk_t new_blocks[4], int *err) 400 { 401 int target, i; 402 unsigned long count = 0; 403 int index = 0; 404 ext2_fsblk_t current_block = 0; 405 int ret = 0; 406 407 /* 408 * Here we try to allocate the requested multiple blocks at once, 409 * on a best-effort basis. 410 * To build a branch, we should allocate blocks for 411 * the indirect blocks(if not allocated yet), and at least 412 * the first direct block of this branch. That's the 413 * minimum number of blocks need to allocate(required) 414 */ 415 target = blks + indirect_blks; 416 417 while (1) { 418 count = target; 419 /* allocating blocks for indirect blocks and direct blocks */ 420 current_block = ext2_new_blocks(inode, goal, &count, err, 0); 421 if (*err) 422 goto failed_out; 423 424 target -= count; 425 /* allocate blocks for indirect blocks */ 426 while (index < indirect_blks && count) { 427 new_blocks[index++] = current_block++; 428 count--; 429 } 430 431 if (count > 0) 432 break; 433 } 434 435 /* save the new block number for the first direct block */ 436 new_blocks[index] = current_block; 437 438 /* total number of blocks allocated for direct blocks */ 439 ret = count; 440 *err = 0; 441 return ret; 442 failed_out: 443 for (i = 0; i <index; i++) 444 ext2_free_blocks(inode, new_blocks[i], 1); 445 if (index) 446 mark_inode_dirty(inode); 447 return ret; 448 } 449 450 /** 451 * ext2_alloc_branch - allocate and set up a chain of blocks. 452 * @inode: owner 453 * @indirect_blks: depth of the chain (number of blocks to allocate) 454 * @blks: number of allocated direct blocks 455 * @goal: preferred place for allocation 456 * @offsets: offsets (in the blocks) to store the pointers to next. 457 * @branch: place to store the chain in. 458 * 459 * This function allocates @num blocks, zeroes out all but the last one, 460 * links them into chain and (if we are synchronous) writes them to disk. 461 * In other words, it prepares a branch that can be spliced onto the 462 * inode. It stores the information about that chain in the branch[], in 463 * the same format as ext2_get_branch() would do. We are calling it after 464 * we had read the existing part of chain and partial points to the last 465 * triple of that (one with zero ->key). Upon the exit we have the same 466 * picture as after the successful ext2_get_block(), except that in one 467 * place chain is disconnected - *branch->p is still zero (we did not 468 * set the last link), but branch->key contains the number that should 469 * be placed into *branch->p to fill that gap. 470 * 471 * If allocation fails we free all blocks we've allocated (and forget 472 * their buffer_heads) and return the error value the from failed 473 * ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain 474 * as described above and return 0. 475 */ 476 477 static int ext2_alloc_branch(struct inode *inode, 478 int indirect_blks, int *blks, ext2_fsblk_t goal, 479 int *offsets, Indirect *branch) 480 { 481 int blocksize = inode->i_sb->s_blocksize; 482 int i, n = 0; 483 int err = 0; 484 struct buffer_head *bh; 485 int num; 486 ext2_fsblk_t new_blocks[4]; 487 ext2_fsblk_t current_block; 488 489 num = ext2_alloc_blocks(inode, goal, indirect_blks, 490 *blks, new_blocks, &err); 491 if (err) 492 return err; 493 494 branch[0].key = cpu_to_le32(new_blocks[0]); 495 /* 496 * metadata blocks and data blocks are allocated. 497 */ 498 for (n = 1; n <= indirect_blks; n++) { 499 /* 500 * Get buffer_head for parent block, zero it out 501 * and set the pointer to new one, then send 502 * parent to disk. 503 */ 504 bh = sb_getblk(inode->i_sb, new_blocks[n-1]); 505 if (unlikely(!bh)) { 506 err = -ENOMEM; 507 goto failed; 508 } 509 branch[n].bh = bh; 510 lock_buffer(bh); 511 memset(bh->b_data, 0, blocksize); 512 branch[n].p = (__le32 *) bh->b_data + offsets[n]; 513 branch[n].key = cpu_to_le32(new_blocks[n]); 514 *branch[n].p = branch[n].key; 515 if ( n == indirect_blks) { 516 current_block = new_blocks[n]; 517 /* 518 * End of chain, update the last new metablock of 519 * the chain to point to the new allocated 520 * data blocks numbers 521 */ 522 for (i=1; i < num; i++) 523 *(branch[n].p + i) = cpu_to_le32(++current_block); 524 } 525 set_buffer_uptodate(bh); 526 unlock_buffer(bh); 527 mmb_mark_buffer_dirty(bh, &EXT2_I(inode)->i_metadata_bhs); 528 /* We used to sync bh here if IS_SYNC(inode). 529 * But we now rely upon generic_write_sync() 530 * and b_inode_buffers. But not for directories. 531 */ 532 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) 533 sync_dirty_buffer(bh); 534 } 535 *blks = num; 536 return err; 537 538 failed: 539 for (i = 1; i < n; i++) 540 bforget(branch[i].bh); 541 for (i = 0; i < indirect_blks; i++) 542 ext2_free_blocks(inode, new_blocks[i], 1); 543 ext2_free_blocks(inode, new_blocks[i], num); 544 return err; 545 } 546 547 /** 548 * ext2_splice_branch - splice the allocated branch onto inode. 549 * @inode: owner 550 * @block: (logical) number of block we are adding 551 * @where: location of missing link 552 * @num: number of indirect blocks we are adding 553 * @blks: number of direct blocks we are adding 554 * 555 * This function fills the missing link and does all housekeeping needed in 556 * inode (->i_blocks, etc.). In case of success we end up with the full 557 * chain to new block and return 0. 558 */ 559 static void ext2_splice_branch(struct inode *inode, 560 long block, Indirect *where, int num, int blks) 561 { 562 int i; 563 struct ext2_block_alloc_info *block_i; 564 ext2_fsblk_t current_block; 565 566 block_i = EXT2_I(inode)->i_block_alloc_info; 567 568 /* XXX LOCKING probably should have i_meta_lock ?*/ 569 /* That's it */ 570 571 *where->p = where->key; 572 573 /* 574 * Update the host buffer_head or inode to point to more just allocated 575 * direct blocks blocks 576 */ 577 if (num == 0 && blks > 1) { 578 current_block = le32_to_cpu(where->key) + 1; 579 for (i = 1; i < blks; i++) 580 *(where->p + i ) = cpu_to_le32(current_block++); 581 } 582 583 /* 584 * update the most recently allocated logical & physical block 585 * in i_block_alloc_info, to assist find the proper goal block for next 586 * allocation 587 */ 588 if (block_i) { 589 block_i->last_alloc_logical_block = block + blks - 1; 590 block_i->last_alloc_physical_block = 591 le32_to_cpu(where[num].key) + blks - 1; 592 } 593 594 /* We are done with atomic stuff, now do the rest of housekeeping */ 595 596 /* had we spliced it onto indirect block? */ 597 if (where->bh) 598 mmb_mark_buffer_dirty(where->bh, &EXT2_I(inode)->i_metadata_bhs); 599 600 inode_set_ctime_current(inode); 601 mark_inode_dirty(inode); 602 } 603 604 /* 605 * Allocation strategy is simple: if we have to allocate something, we will 606 * have to go the whole way to leaf. So let's do it before attaching anything 607 * to tree, set linkage between the newborn blocks, write them if sync is 608 * required, recheck the path, free and repeat if check fails, otherwise 609 * set the last missing link (that will protect us from any truncate-generated 610 * removals - all blocks on the path are immune now) and possibly force the 611 * write on the parent block. 612 * That has a nice additional property: no special recovery from the failed 613 * allocations is needed - we simply release blocks and do not touch anything 614 * reachable from inode. 615 * 616 * `handle' can be NULL if create == 0. 617 * 618 * return > 0, # of blocks mapped or allocated. 619 * return = 0, if plain lookup failed. 620 * return < 0, error case. 621 */ 622 static int ext2_get_blocks(struct inode *inode, 623 sector_t iblock, unsigned long maxblocks, 624 u32 *bno, bool *new, bool *boundary, 625 int create) 626 { 627 int err; 628 int offsets[4]; 629 Indirect chain[4]; 630 Indirect *partial; 631 ext2_fsblk_t goal; 632 int indirect_blks; 633 int blocks_to_boundary = 0; 634 int depth; 635 struct ext2_inode_info *ei = EXT2_I(inode); 636 int count = 0; 637 ext2_fsblk_t first_block = 0; 638 639 if (WARN_ON_ONCE(maxblocks == 0)) 640 return -EINVAL; 641 642 depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary); 643 644 if (depth == 0) 645 return -EIO; 646 647 partial = ext2_get_branch(inode, depth, offsets, chain, &err); 648 /* Simplest case - block found, no allocation needed */ 649 if (!partial) { 650 first_block = le32_to_cpu(chain[depth - 1].key); 651 count++; 652 /*map more blocks*/ 653 while (count < maxblocks && count <= blocks_to_boundary) { 654 ext2_fsblk_t blk; 655 656 if (!verify_chain(chain, chain + depth - 1)) { 657 /* 658 * Indirect block might be removed by 659 * truncate while we were reading it. 660 * Handling of that case: forget what we've 661 * got now, go to reread. 662 */ 663 err = -EAGAIN; 664 count = 0; 665 partial = chain + depth - 1; 666 break; 667 } 668 blk = le32_to_cpu(*(chain[depth-1].p + count)); 669 if (blk == first_block + count) 670 count++; 671 else 672 break; 673 } 674 if (err != -EAGAIN) 675 goto got_it; 676 } 677 678 /* Next simple case - plain lookup or failed read of indirect block */ 679 if (!create || err == -EIO) 680 goto cleanup; 681 682 mutex_lock(&ei->truncate_mutex); 683 /* 684 * If the indirect block is missing while we are reading 685 * the chain(ext2_get_branch() returns -EAGAIN err), or 686 * if the chain has been changed after we grab the semaphore, 687 * (either because another process truncated this branch, or 688 * another get_block allocated this branch) re-grab the chain to see if 689 * the request block has been allocated or not. 690 * 691 * Since we already block the truncate/other get_block 692 * at this point, we will have the current copy of the chain when we 693 * splice the branch into the tree. 694 */ 695 if (err == -EAGAIN || !verify_chain(chain, partial)) { 696 while (partial > chain) { 697 brelse(partial->bh); 698 partial--; 699 } 700 partial = ext2_get_branch(inode, depth, offsets, chain, &err); 701 if (!partial) { 702 count++; 703 mutex_unlock(&ei->truncate_mutex); 704 goto got_it; 705 } 706 707 if (err) { 708 mutex_unlock(&ei->truncate_mutex); 709 goto cleanup; 710 } 711 } 712 713 /* 714 * Okay, we need to do block allocation. Lazily initialize the block 715 * allocation info here if necessary 716 */ 717 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info)) 718 ext2_init_block_alloc_info(inode); 719 720 goal = ext2_find_goal(inode, iblock, partial); 721 722 /* the number of blocks need to allocate for [d,t]indirect blocks */ 723 indirect_blks = (chain + depth) - partial - 1; 724 /* 725 * Next look up the indirect map to count the total number of 726 * direct blocks to allocate for this branch. 727 */ 728 count = ext2_blks_to_allocate(partial, indirect_blks, 729 maxblocks, blocks_to_boundary); 730 /* 731 * XXX ???? Block out ext2_truncate while we alter the tree 732 */ 733 err = ext2_alloc_branch(inode, indirect_blks, &count, goal, 734 offsets + (partial - chain), partial); 735 736 if (err) { 737 mutex_unlock(&ei->truncate_mutex); 738 goto cleanup; 739 } 740 741 *new = true; 742 743 ext2_splice_branch(inode, iblock, partial, indirect_blks, count); 744 mutex_unlock(&ei->truncate_mutex); 745 got_it: 746 if (count > blocks_to_boundary) 747 *boundary = true; 748 err = count; 749 /* Clean up and exit */ 750 partial = chain + depth - 1; /* the whole chain */ 751 cleanup: 752 while (partial > chain) { 753 brelse(partial->bh); 754 partial--; 755 } 756 if (err > 0) 757 *bno = le32_to_cpu(chain[depth-1].key); 758 return err; 759 } 760 761 int ext2_get_block(struct inode *inode, sector_t iblock, 762 struct buffer_head *bh_result, int create) 763 { 764 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; 765 bool new = false, boundary = false; 766 u32 bno; 767 int ret; 768 769 ret = ext2_get_blocks(inode, iblock, max_blocks, &bno, &new, &boundary, 770 create); 771 if (ret <= 0) 772 return ret; 773 774 map_bh(bh_result, inode->i_sb, bno); 775 bh_result->b_size = (ret << inode->i_blkbits); 776 if (new) 777 set_buffer_new(bh_result); 778 if (boundary) 779 set_buffer_boundary(bh_result); 780 return 0; 781 782 } 783 784 static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length, 785 unsigned flags, struct iomap *iomap, struct iomap *srcmap) 786 { 787 unsigned int blkbits = inode->i_blkbits; 788 unsigned long first_block = offset >> blkbits; 789 unsigned long max_blocks = (length + (1 << blkbits) - 1) >> blkbits; 790 bool new = false, boundary = false; 791 u32 bno; 792 int ret; 793 bool create = flags & IOMAP_WRITE; 794 795 /* 796 * For writes that could fill holes inside i_size on a 797 * DIO_SKIP_HOLES filesystem we forbid block creations: only 798 * overwrites are permitted. 799 */ 800 if ((flags & IOMAP_DIRECT) && 801 (first_block << blkbits) < i_size_read(inode)) 802 create = 0; 803 804 /* 805 * Writes that span EOF might trigger an IO size update on completion, 806 * so consider them to be dirty for the purposes of O_DSYNC even if 807 * there is no other metadata changes pending or have been made here. 808 */ 809 if ((flags & IOMAP_WRITE) && offset + length > i_size_read(inode)) 810 iomap->flags |= IOMAP_F_DIRTY; 811 812 ret = ext2_get_blocks(inode, first_block, max_blocks, 813 &bno, &new, &boundary, create); 814 if (ret < 0) 815 return ret; 816 817 iomap->flags = 0; 818 iomap->offset = (u64)first_block << blkbits; 819 iomap->bdev = inode->i_sb->s_bdev; 820 821 if (ret == 0) { 822 /* 823 * Switch to buffered-io for writing to holes in a non-extent 824 * based filesystem to avoid stale data exposure problem. 825 */ 826 if (!create && (flags & IOMAP_WRITE) && (flags & IOMAP_DIRECT)) 827 return -ENOTBLK; 828 iomap->type = IOMAP_HOLE; 829 iomap->addr = IOMAP_NULL_ADDR; 830 iomap->length = 1 << blkbits; 831 } else { 832 iomap->type = IOMAP_MAPPED; 833 iomap->addr = (u64)bno << blkbits; 834 iomap->length = (u64)ret << blkbits; 835 iomap->flags |= IOMAP_F_MERGED; 836 } 837 838 if (new) 839 iomap->flags |= IOMAP_F_NEW; 840 return 0; 841 } 842 843 static int 844 ext2_iomap_end(struct inode *inode, loff_t offset, loff_t length, 845 ssize_t written, unsigned flags, struct iomap *iomap) 846 { 847 /* 848 * Switch to buffered-io in case of any error. 849 * Blocks allocated can be used by the buffered-io path. 850 */ 851 if ((flags & IOMAP_DIRECT) && (flags & IOMAP_WRITE) && written == 0) 852 return -ENOTBLK; 853 854 if (iomap->type == IOMAP_MAPPED && 855 written < length && 856 (flags & IOMAP_WRITE)) 857 ext2_write_failed(inode->i_mapping, offset + length); 858 return 0; 859 } 860 861 static DEFINE_IOMAP_ITER_NEXT_END(ext2_iomap_next, ext2_iomap_begin, 862 ext2_iomap_end); 863 864 const struct iomap_ops ext2_iomap_ops = { 865 .iomap_next = ext2_iomap_next, 866 }; 867 868 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 869 u64 start, u64 len) 870 { 871 int ret; 872 loff_t i_size; 873 874 inode_lock(inode); 875 i_size = i_size_read(inode); 876 /* 877 * iomap_fiemap() returns EINVAL for 0 length. Make sure we don't trim 878 * length to 0 but still trim the range as much as possible since 879 * ext2_get_blocks() iterates unmapped space block by block which is 880 * slow. 881 */ 882 if (i_size == 0) 883 i_size = 1; 884 len = min_t(u64, len, i_size); 885 ret = iomap_fiemap(inode, fieinfo, start, len, &ext2_iomap_ops); 886 inode_unlock(inode); 887 888 return ret; 889 } 890 891 static int ext2_read_folio(struct file *file, struct folio *folio) 892 { 893 return mpage_read_folio(folio, ext2_get_block); 894 } 895 896 static void ext2_readahead(struct readahead_control *rac) 897 { 898 mpage_readahead(rac, ext2_get_block); 899 } 900 901 static int 902 ext2_write_begin(const struct kiocb *iocb, struct address_space *mapping, 903 loff_t pos, unsigned len, struct folio **foliop, void **fsdata) 904 { 905 int ret; 906 907 ret = block_write_begin(mapping, pos, len, foliop, ext2_get_block); 908 if (ret < 0) 909 ext2_write_failed(mapping, pos + len); 910 return ret; 911 } 912 913 static int ext2_write_end(const struct kiocb *iocb, 914 struct address_space *mapping, 915 loff_t pos, unsigned len, unsigned copied, 916 struct folio *folio, void *fsdata) 917 { 918 int ret; 919 920 ret = generic_write_end(iocb, mapping, pos, len, copied, folio, fsdata); 921 if (ret < len) 922 ext2_write_failed(mapping, pos + len); 923 return ret; 924 } 925 926 static sector_t ext2_bmap(struct address_space *mapping, sector_t block) 927 { 928 return generic_block_bmap(mapping,block,ext2_get_block); 929 } 930 931 static int 932 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc) 933 { 934 return mpage_writepages(mapping, wbc, ext2_get_block); 935 } 936 937 938 const struct address_space_operations ext2_aops = { 939 .dirty_folio = block_dirty_folio, 940 .invalidate_folio = block_invalidate_folio, 941 .read_folio = ext2_read_folio, 942 .readahead = ext2_readahead, 943 .write_begin = ext2_write_begin, 944 .write_end = ext2_write_end, 945 .bmap = ext2_bmap, 946 .writepages = ext2_writepages, 947 .migrate_folio = buffer_migrate_folio, 948 .is_partially_uptodate = block_is_partially_uptodate, 949 .error_remove_folio = generic_error_remove_folio, 950 }; 951 952 953 /* 954 * Probably it should be a library function... search for first non-zero word 955 * or memcmp with zero_page, whatever is better for particular architecture. 956 * Linus? 957 */ 958 static inline int all_zeroes(__le32 *p, __le32 *q) 959 { 960 while (p < q) 961 if (*p++) 962 return 0; 963 return 1; 964 } 965 966 /** 967 * ext2_find_shared - find the indirect blocks for partial truncation. 968 * @inode: inode in question 969 * @depth: depth of the affected branch 970 * @offsets: offsets of pointers in that branch (see ext2_block_to_path) 971 * @chain: place to store the pointers to partial indirect blocks 972 * @top: place to the (detached) top of branch 973 * 974 * This is a helper function used by ext2_truncate(). 975 * 976 * When we do truncate() we may have to clean the ends of several indirect 977 * blocks but leave the blocks themselves alive. Block is partially 978 * truncated if some data below the new i_size is referred from it (and 979 * it is on the path to the first completely truncated data block, indeed). 980 * We have to free the top of that path along with everything to the right 981 * of the path. Since no allocation past the truncation point is possible 982 * until ext2_truncate() finishes, we may safely do the latter, but top 983 * of branch may require special attention - pageout below the truncation 984 * point might try to populate it. 985 * 986 * We atomically detach the top of branch from the tree, store the block 987 * number of its root in *@top, pointers to buffer_heads of partially 988 * truncated blocks - in @chain[].bh and pointers to their last elements 989 * that should not be removed - in @chain[].p. Return value is the pointer 990 * to last filled element of @chain. 991 * 992 * The work left to caller to do the actual freeing of subtrees: 993 * a) free the subtree starting from *@top 994 * b) free the subtrees whose roots are stored in 995 * (@chain[i].p+1 .. end of @chain[i].bh->b_data) 996 * c) free the subtrees growing from the inode past the @chain[0].p 997 * (no partially truncated stuff there). 998 */ 999 1000 static Indirect *ext2_find_shared(struct inode *inode, 1001 int depth, 1002 int offsets[4], 1003 Indirect chain[4], 1004 __le32 *top) 1005 { 1006 Indirect *partial, *p; 1007 int k, err; 1008 1009 *top = 0; 1010 for (k = depth; k > 1 && !offsets[k-1]; k--) 1011 ; 1012 partial = ext2_get_branch(inode, k, offsets, chain, &err); 1013 if (!partial) 1014 partial = chain + k-1; 1015 /* 1016 * If the branch acquired continuation since we've looked at it - 1017 * fine, it should all survive and (new) top doesn't belong to us. 1018 */ 1019 write_lock(&EXT2_I(inode)->i_meta_lock); 1020 if (!partial->key && *partial->p) { 1021 write_unlock(&EXT2_I(inode)->i_meta_lock); 1022 goto no_top; 1023 } 1024 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--) 1025 ; 1026 /* 1027 * OK, we've found the last block that must survive. The rest of our 1028 * branch should be detached before unlocking. However, if that rest 1029 * of branch is all ours and does not grow immediately from the inode 1030 * it's easier to cheat and just decrement partial->p. 1031 */ 1032 if (p == chain + k - 1 && p > chain) { 1033 p->p--; 1034 } else { 1035 *top = *p->p; 1036 *p->p = 0; 1037 } 1038 write_unlock(&EXT2_I(inode)->i_meta_lock); 1039 1040 while(partial > p) 1041 { 1042 brelse(partial->bh); 1043 partial--; 1044 } 1045 no_top: 1046 return partial; 1047 } 1048 1049 /** 1050 * ext2_free_data - free a list of data blocks 1051 * @inode: inode we are dealing with 1052 * @p: array of block numbers 1053 * @q: points immediately past the end of array 1054 * 1055 * We are freeing all blocks referred from that array (numbers are 1056 * stored as little-endian 32-bit) and updating @inode->i_blocks 1057 * appropriately. 1058 */ 1059 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q) 1060 { 1061 ext2_fsblk_t block_to_free = 0, count = 0; 1062 ext2_fsblk_t nr; 1063 1064 for ( ; p < q ; p++) { 1065 nr = le32_to_cpu(*p); 1066 if (nr) { 1067 *p = 0; 1068 /* accumulate blocks to free if they're contiguous */ 1069 if (count == 0) 1070 goto free_this; 1071 else if (block_to_free == nr - count) 1072 count++; 1073 else { 1074 ext2_free_blocks (inode, block_to_free, count); 1075 mark_inode_dirty(inode); 1076 free_this: 1077 block_to_free = nr; 1078 count = 1; 1079 } 1080 } 1081 } 1082 if (count > 0) { 1083 ext2_free_blocks (inode, block_to_free, count); 1084 mark_inode_dirty(inode); 1085 } 1086 } 1087 1088 /** 1089 * ext2_free_branches - free an array of branches 1090 * @inode: inode we are dealing with 1091 * @p: array of block numbers 1092 * @q: pointer immediately past the end of array 1093 * @depth: depth of the branches to free 1094 * 1095 * We are freeing all blocks referred from these branches (numbers are 1096 * stored as little-endian 32-bit) and updating @inode->i_blocks 1097 * appropriately. 1098 */ 1099 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth) 1100 { 1101 struct buffer_head * bh; 1102 ext2_fsblk_t nr; 1103 1104 if (depth--) { 1105 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb); 1106 for ( ; p < q ; p++) { 1107 nr = le32_to_cpu(*p); 1108 if (!nr) 1109 continue; 1110 *p = 0; 1111 bh = sb_bread(inode->i_sb, nr); 1112 /* 1113 * A read failure? Report error and clear slot 1114 * (should be rare). 1115 */ 1116 if (!bh) { 1117 ext2_error(inode->i_sb, "ext2_free_branches", 1118 "Read failure, inode=%llu, block=%ld", 1119 inode->i_ino, nr); 1120 continue; 1121 } 1122 ext2_free_branches(inode, 1123 (__le32*)bh->b_data, 1124 (__le32*)bh->b_data + addr_per_block, 1125 depth); 1126 bforget(bh); 1127 ext2_free_blocks(inode, nr, 1); 1128 mark_inode_dirty(inode); 1129 } 1130 } else 1131 ext2_free_data(inode, p, q); 1132 } 1133 1134 /* mapping->invalidate_lock must be held when calling this function */ 1135 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset) 1136 { 1137 __le32 *i_data = EXT2_I(inode)->i_data; 1138 struct ext2_inode_info *ei = EXT2_I(inode); 1139 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb); 1140 int offsets[4]; 1141 Indirect chain[4]; 1142 Indirect *partial; 1143 __le32 nr = 0; 1144 int n; 1145 long iblock; 1146 unsigned blocksize; 1147 blocksize = inode->i_sb->s_blocksize; 1148 iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb); 1149 1150 1151 n = ext2_block_to_path(inode, iblock, offsets, NULL); 1152 if (n == 0) 1153 return; 1154 1155 /* 1156 * From here we block out all ext2_get_block() callers who want to 1157 * modify the block allocation tree. 1158 */ 1159 mutex_lock(&ei->truncate_mutex); 1160 1161 if (n == 1) { 1162 ext2_free_data(inode, i_data+offsets[0], 1163 i_data + EXT2_NDIR_BLOCKS); 1164 goto do_indirects; 1165 } 1166 1167 partial = ext2_find_shared(inode, n, offsets, chain, &nr); 1168 /* Kill the top of shared branch (already detached) */ 1169 if (nr) { 1170 if (partial == chain) 1171 mark_inode_dirty(inode); 1172 else 1173 mmb_mark_buffer_dirty(partial->bh, 1174 &EXT2_I(inode)->i_metadata_bhs); 1175 ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial); 1176 } 1177 /* Clear the ends of indirect blocks on the shared branch */ 1178 while (partial > chain) { 1179 ext2_free_branches(inode, 1180 partial->p + 1, 1181 (__le32*)partial->bh->b_data+addr_per_block, 1182 (chain+n-1) - partial); 1183 mmb_mark_buffer_dirty(partial->bh, 1184 &EXT2_I(inode)->i_metadata_bhs); 1185 brelse (partial->bh); 1186 partial--; 1187 } 1188 do_indirects: 1189 /* Kill the remaining (whole) subtrees */ 1190 switch (offsets[0]) { 1191 default: 1192 nr = i_data[EXT2_IND_BLOCK]; 1193 if (nr) { 1194 i_data[EXT2_IND_BLOCK] = 0; 1195 mark_inode_dirty(inode); 1196 ext2_free_branches(inode, &nr, &nr+1, 1); 1197 } 1198 fallthrough; 1199 case EXT2_IND_BLOCK: 1200 nr = i_data[EXT2_DIND_BLOCK]; 1201 if (nr) { 1202 i_data[EXT2_DIND_BLOCK] = 0; 1203 mark_inode_dirty(inode); 1204 ext2_free_branches(inode, &nr, &nr+1, 2); 1205 } 1206 fallthrough; 1207 case EXT2_DIND_BLOCK: 1208 nr = i_data[EXT2_TIND_BLOCK]; 1209 if (nr) { 1210 i_data[EXT2_TIND_BLOCK] = 0; 1211 mark_inode_dirty(inode); 1212 ext2_free_branches(inode, &nr, &nr+1, 3); 1213 } 1214 break; 1215 case EXT2_TIND_BLOCK: 1216 ; 1217 } 1218 1219 ext2_discard_reservation(inode); 1220 1221 mutex_unlock(&ei->truncate_mutex); 1222 } 1223 1224 static void ext2_truncate_blocks(struct inode *inode, loff_t offset) 1225 { 1226 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1227 S_ISLNK(inode->i_mode))) 1228 return; 1229 if (ext2_inode_is_fast_symlink(inode)) 1230 return; 1231 1232 filemap_invalidate_lock(inode->i_mapping); 1233 __ext2_truncate_blocks(inode, offset); 1234 filemap_invalidate_unlock(inode->i_mapping); 1235 } 1236 1237 static int ext2_setsize(struct inode *inode, loff_t newsize) 1238 { 1239 int error; 1240 1241 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1242 S_ISLNK(inode->i_mode))) 1243 return -EINVAL; 1244 if (ext2_inode_is_fast_symlink(inode)) 1245 return -EINVAL; 1246 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 1247 return -EPERM; 1248 1249 inode_dio_wait(inode); 1250 1251 error = block_truncate_page(inode->i_mapping, newsize, ext2_get_block); 1252 if (error) 1253 return error; 1254 1255 filemap_invalidate_lock(inode->i_mapping); 1256 truncate_setsize(inode, newsize); 1257 __ext2_truncate_blocks(inode, newsize); 1258 filemap_invalidate_unlock(inode->i_mapping); 1259 1260 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); 1261 mark_inode_dirty(inode); 1262 if (inode_needs_sync(inode)) 1263 sync_inode_metadata(inode, 1); 1264 1265 return 0; 1266 } 1267 1268 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino, 1269 struct buffer_head **p) 1270 { 1271 struct buffer_head * bh; 1272 unsigned long block_group; 1273 unsigned long block; 1274 unsigned long offset; 1275 struct ext2_group_desc * gdp; 1276 1277 *p = NULL; 1278 if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) || 1279 ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count)) 1280 goto Einval; 1281 1282 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb); 1283 gdp = ext2_get_group_desc(sb, block_group, NULL); 1284 if (!gdp) 1285 goto Egdp; 1286 /* 1287 * Figure out the offset within the block group inode table 1288 */ 1289 offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb); 1290 block = le32_to_cpu(gdp->bg_inode_table) + 1291 (offset >> EXT2_BLOCK_SIZE_BITS(sb)); 1292 if (!(bh = sb_bread(sb, block))) 1293 goto Eio; 1294 1295 *p = bh; 1296 offset &= (EXT2_BLOCK_SIZE(sb) - 1); 1297 return (struct ext2_inode *) (bh->b_data + offset); 1298 1299 Einval: 1300 ext2_error(sb, "ext2_get_inode", "bad inode number: %lu", 1301 (unsigned long) ino); 1302 return ERR_PTR(-EINVAL); 1303 Eio: 1304 ext2_error(sb, "ext2_get_inode", 1305 "unable to read inode block - inode=%lu, block=%lu", 1306 (unsigned long) ino, block); 1307 Egdp: 1308 return ERR_PTR(-EIO); 1309 } 1310 1311 void ext2_set_inode_flags(struct inode *inode) 1312 { 1313 unsigned int flags = EXT2_I(inode)->i_flags; 1314 1315 inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | 1316 S_DIRSYNC); 1317 if (flags & EXT2_SYNC_FL) 1318 inode->i_flags |= S_SYNC; 1319 if (flags & EXT2_APPEND_FL) 1320 inode->i_flags |= S_APPEND; 1321 if (flags & EXT2_IMMUTABLE_FL) 1322 inode->i_flags |= S_IMMUTABLE; 1323 if (flags & EXT2_NOATIME_FL) 1324 inode->i_flags |= S_NOATIME; 1325 if (flags & EXT2_DIRSYNC_FL) 1326 inode->i_flags |= S_DIRSYNC; 1327 } 1328 1329 void ext2_set_file_ops(struct inode *inode) 1330 { 1331 inode->i_op = &ext2_file_inode_operations; 1332 inode->i_fop = &ext2_file_operations; 1333 inode->i_mapping->a_ops = &ext2_aops; 1334 } 1335 1336 struct inode *ext2_iget (struct super_block *sb, unsigned long ino) 1337 { 1338 struct ext2_inode_info *ei; 1339 struct buffer_head * bh = NULL; 1340 struct ext2_inode *raw_inode; 1341 struct inode *inode; 1342 long ret = -EIO; 1343 int n; 1344 uid_t i_uid; 1345 gid_t i_gid; 1346 1347 inode = iget_locked(sb, ino); 1348 if (!inode) 1349 return ERR_PTR(-ENOMEM); 1350 if (!(inode_state_read_once(inode) & I_NEW)) 1351 return inode; 1352 1353 ei = EXT2_I(inode); 1354 ei->i_block_alloc_info = NULL; 1355 1356 raw_inode = ext2_get_inode(inode->i_sb, ino, &bh); 1357 if (IS_ERR(raw_inode)) { 1358 ret = PTR_ERR(raw_inode); 1359 goto bad_inode; 1360 } 1361 1362 inode->i_mode = le16_to_cpu(raw_inode->i_mode); 1363 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); 1364 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); 1365 if (!(test_opt (inode->i_sb, NO_UID32))) { 1366 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; 1367 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; 1368 } 1369 i_uid_write(inode, i_uid); 1370 i_gid_write(inode, i_gid); 1371 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); 1372 inode->i_size = le32_to_cpu(raw_inode->i_size); 1373 inode_set_atime(inode, (signed)le32_to_cpu(raw_inode->i_atime), 0); 1374 inode_set_ctime(inode, (signed)le32_to_cpu(raw_inode->i_ctime), 0); 1375 inode_set_mtime(inode, (signed)le32_to_cpu(raw_inode->i_mtime), 0); 1376 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); 1377 /* We now have enough fields to check if the inode was active or not. 1378 * This is needed because nfsd might try to access dead inodes 1379 * the test is that same one that e2fsck uses 1380 * NeilBrown 1999oct15 1381 */ 1382 if (inode->i_nlink == 0) { 1383 if (inode->i_mode == 0 || ei->i_dtime) { 1384 /* this inode is deleted */ 1385 ret = -ESTALE; 1386 } else { 1387 ext2_error(sb, __func__, 1388 "inode %lu has zero i_nlink with mode 0%o and no dtime, " 1389 "filesystem may be corrupt", 1390 ino, inode->i_mode); 1391 ret = -EFSCORRUPTED; 1392 } 1393 goto bad_inode; 1394 } 1395 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks); 1396 ei->i_flags = le32_to_cpu(raw_inode->i_flags); 1397 ext2_set_inode_flags(inode); 1398 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr); 1399 ei->i_frag_no = raw_inode->i_frag; 1400 ei->i_frag_size = raw_inode->i_fsize; 1401 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl); 1402 ei->i_dir_acl = 0; 1403 1404 if (ei->i_file_acl && 1405 !ext2_data_block_valid(EXT2_SB(sb), ei->i_file_acl, 1)) { 1406 ext2_error(sb, "ext2_iget", "bad extended attribute block %u", 1407 ei->i_file_acl); 1408 ret = -EFSCORRUPTED; 1409 goto bad_inode; 1410 } 1411 1412 if (S_ISREG(inode->i_mode)) 1413 inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32; 1414 else 1415 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl); 1416 if (i_size_read(inode) < 0) { 1417 ret = -EFSCORRUPTED; 1418 goto bad_inode; 1419 } 1420 ei->i_dtime = 0; 1421 inode->i_generation = le32_to_cpu(raw_inode->i_generation); 1422 ei->i_state = 0; 1423 ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb); 1424 ei->i_dir_start_lookup = 0; 1425 1426 /* 1427 * NOTE! The in-memory inode i_data array is in little-endian order 1428 * even on big-endian machines: we do NOT byteswap the block numbers! 1429 */ 1430 for (n = 0; n < EXT2_N_BLOCKS; n++) 1431 ei->i_data[n] = raw_inode->i_block[n]; 1432 1433 if (S_ISREG(inode->i_mode)) { 1434 ext2_set_file_ops(inode); 1435 } else if (S_ISDIR(inode->i_mode)) { 1436 inode->i_op = &ext2_dir_inode_operations; 1437 inode->i_fop = &ext2_dir_operations; 1438 inode->i_mapping->a_ops = &ext2_aops; 1439 } else if (S_ISLNK(inode->i_mode)) { 1440 if (ext2_inode_is_fast_symlink(inode)) { 1441 inode->i_link = (char *)ei->i_data; 1442 inode->i_op = &ext2_fast_symlink_inode_operations; 1443 nd_terminate_link(ei->i_data, inode->i_size, 1444 sizeof(ei->i_data) - 1); 1445 } else { 1446 inode->i_op = &ext2_symlink_inode_operations; 1447 inode_nohighmem(inode); 1448 inode->i_mapping->a_ops = &ext2_aops; 1449 } 1450 } else { 1451 inode->i_op = &ext2_special_inode_operations; 1452 if (raw_inode->i_block[0]) 1453 init_special_inode(inode, inode->i_mode, 1454 old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); 1455 else 1456 init_special_inode(inode, inode->i_mode, 1457 new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); 1458 } 1459 brelse (bh); 1460 unlock_new_inode(inode); 1461 return inode; 1462 1463 bad_inode: 1464 brelse(bh); 1465 iget_failed(inode); 1466 return ERR_PTR(ret); 1467 } 1468 1469 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc) 1470 { 1471 struct ext2_inode_info *ei = EXT2_I(inode); 1472 struct super_block *sb = inode->i_sb; 1473 ino_t ino = inode->i_ino; 1474 uid_t uid = i_uid_read(inode); 1475 gid_t gid = i_gid_read(inode); 1476 struct buffer_head * bh; 1477 struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh); 1478 int n; 1479 int err = 0; 1480 1481 if (IS_ERR(raw_inode)) 1482 return -EIO; 1483 1484 /* For fields not tracking in the in-memory inode, 1485 * initialise them to zero for new inodes. */ 1486 if (ei->i_state & EXT2_STATE_NEW) 1487 memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size); 1488 1489 raw_inode->i_mode = cpu_to_le16(inode->i_mode); 1490 if (!(test_opt(sb, NO_UID32))) { 1491 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid)); 1492 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid)); 1493 /* 1494 * Fix up interoperability with old kernels. Otherwise, old inodes get 1495 * re-used with the upper 16 bits of the uid/gid intact 1496 */ 1497 if (!ei->i_dtime) { 1498 raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid)); 1499 raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid)); 1500 } else { 1501 raw_inode->i_uid_high = 0; 1502 raw_inode->i_gid_high = 0; 1503 } 1504 } else { 1505 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid)); 1506 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid)); 1507 raw_inode->i_uid_high = 0; 1508 raw_inode->i_gid_high = 0; 1509 } 1510 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); 1511 raw_inode->i_size = cpu_to_le32(inode->i_size); 1512 raw_inode->i_atime = cpu_to_le32(inode_get_atime_sec(inode)); 1513 raw_inode->i_ctime = cpu_to_le32(inode_get_ctime_sec(inode)); 1514 raw_inode->i_mtime = cpu_to_le32(inode_get_mtime_sec(inode)); 1515 1516 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks); 1517 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); 1518 raw_inode->i_flags = cpu_to_le32(ei->i_flags); 1519 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr); 1520 raw_inode->i_frag = ei->i_frag_no; 1521 raw_inode->i_fsize = ei->i_frag_size; 1522 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl); 1523 if (!S_ISREG(inode->i_mode)) 1524 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl); 1525 else { 1526 raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32); 1527 if (inode->i_size > 0x7fffffffULL) { 1528 if (!EXT2_HAS_RO_COMPAT_FEATURE(sb, 1529 EXT2_FEATURE_RO_COMPAT_LARGE_FILE) || 1530 EXT2_SB(sb)->s_es->s_rev_level == 1531 cpu_to_le32(EXT2_GOOD_OLD_REV)) { 1532 /* If this is the first large file 1533 * created, add a flag to the superblock. 1534 */ 1535 spin_lock(&EXT2_SB(sb)->s_lock); 1536 ext2_update_dynamic_rev(sb); 1537 EXT2_SET_RO_COMPAT_FEATURE(sb, 1538 EXT2_FEATURE_RO_COMPAT_LARGE_FILE); 1539 spin_unlock(&EXT2_SB(sb)->s_lock); 1540 ext2_sync_super(sb, EXT2_SB(sb)->s_es, 1); 1541 } 1542 } 1543 } 1544 1545 raw_inode->i_generation = cpu_to_le32(inode->i_generation); 1546 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { 1547 if (old_valid_dev(inode->i_rdev)) { 1548 raw_inode->i_block[0] = 1549 cpu_to_le32(old_encode_dev(inode->i_rdev)); 1550 raw_inode->i_block[1] = 0; 1551 } else { 1552 raw_inode->i_block[0] = 0; 1553 raw_inode->i_block[1] = 1554 cpu_to_le32(new_encode_dev(inode->i_rdev)); 1555 raw_inode->i_block[2] = 0; 1556 } 1557 } else for (n = 0; n < EXT2_N_BLOCKS; n++) 1558 raw_inode->i_block[n] = ei->i_data[n]; 1559 mark_buffer_dirty(bh); 1560 ei->i_state &= ~EXT2_STATE_NEW; 1561 brelse (bh); 1562 set_inode_metadata_writeback(inode); 1563 return err; 1564 } 1565 1566 int ext2_sync_inode_metadata(struct inode *inode, struct writeback_control *wbc) 1567 { 1568 struct buffer_head *bh; 1569 struct ext2_inode *raw_inode = ext2_get_inode(inode->i_sb, inode->i_ino, 1570 &bh); 1571 int err = 0; 1572 1573 if (IS_ERR(raw_inode)) 1574 return -EIO; 1575 err = mmb_sync(&EXT2_I(inode)->i_metadata_bhs); 1576 if (err) { 1577 ext2_error(inode->i_sb, __func__, 1578 "Error syncing inode metadata ino=%lu\n", 1579 (unsigned long)inode->i_ino); 1580 goto out; 1581 } 1582 sync_dirty_buffer(bh); 1583 if (buffer_write_io_error(bh)) { 1584 ext2_error(inode->i_sb, __func__, 1585 "IO error syncing inode %lu\n", 1586 (unsigned long)inode->i_ino); 1587 err = -EIO; 1588 } 1589 out: 1590 brelse(bh); 1591 return err; 1592 } 1593 1594 int ext2_getattr(struct mnt_idmap *idmap, const struct path *path, 1595 struct kstat *stat, u32 request_mask, unsigned int query_flags) 1596 { 1597 struct inode *inode = d_inode(path->dentry); 1598 struct ext2_inode_info *ei = EXT2_I(inode); 1599 unsigned int flags; 1600 1601 flags = ei->i_flags & EXT2_FL_USER_VISIBLE; 1602 if (flags & EXT2_APPEND_FL) 1603 stat->attributes |= STATX_ATTR_APPEND; 1604 if (flags & EXT2_COMPR_FL) 1605 stat->attributes |= STATX_ATTR_COMPRESSED; 1606 if (flags & EXT2_IMMUTABLE_FL) 1607 stat->attributes |= STATX_ATTR_IMMUTABLE; 1608 if (flags & EXT2_NODUMP_FL) 1609 stat->attributes |= STATX_ATTR_NODUMP; 1610 stat->attributes_mask |= (STATX_ATTR_APPEND | 1611 STATX_ATTR_COMPRESSED | 1612 STATX_ATTR_ENCRYPTED | 1613 STATX_ATTR_IMMUTABLE | 1614 STATX_ATTR_NODUMP); 1615 1616 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat); 1617 return 0; 1618 } 1619 1620 int ext2_setattr(struct mnt_idmap *idmap, struct dentry *dentry, 1621 struct iattr *iattr) 1622 { 1623 struct inode *inode = d_inode(dentry); 1624 int error; 1625 1626 error = setattr_prepare(&nop_mnt_idmap, dentry, iattr); 1627 if (error) 1628 return error; 1629 1630 if (is_quota_modification(&nop_mnt_idmap, inode, iattr)) { 1631 error = dquot_initialize(inode); 1632 if (error) 1633 return error; 1634 } 1635 if (i_uid_needs_update(&nop_mnt_idmap, iattr, inode) || 1636 i_gid_needs_update(&nop_mnt_idmap, iattr, inode)) { 1637 error = dquot_transfer(&nop_mnt_idmap, inode, iattr); 1638 if (error) 1639 return error; 1640 } 1641 if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) { 1642 error = ext2_setsize(inode, iattr->ia_size); 1643 if (error) 1644 return error; 1645 } 1646 setattr_copy(&nop_mnt_idmap, inode, iattr); 1647 if (iattr->ia_valid & ATTR_MODE) 1648 error = posix_acl_chmod(&nop_mnt_idmap, dentry, inode->i_mode); 1649 mark_inode_dirty(inode); 1650 1651 return error; 1652 } 1653